Carl E. Landwehr. Purdue University. West Lafayette, Indiana 47907 i. Introduction. The following report is presented in response to Professor Browne's request.
Usage Statistics For MTS Carl E. Landwehr Purdue University West Lafayette, Indiana 47907
i.
Introduction The following report is presented in response to Professor Browne's request for case studies of performance measurement projects; this study takes a macroscopic view of a large-scale time sharing and batch processing installation. The data from the University of Michigan Computing Center presented in this paper were gathered for the purpose of constructing and validating an analytical model for a host system in a computer network (i). Since a computer network consists of several host systems, a simple model for a single host was required in order to keep the network model mathematically tractable. Consequently, measurement efforts were focussed on the limiting resource in the system, the CPU. Also, since waiting time statistics for batch jobs were not available, measurements of batch queue lengths were made in order to assess system performance in relation to batch jobs.
2.
Data Sources Considerable quantities of data are collected routinely by the University of Michigan Terminal System. The data presented in this paper come from two sources: job accounting statistics (recorded on a per-job basis) and system status measurements (recorded on a per-unit-time basis). Ultimate responsibility for the generation of these records lies with the supervisor. Both accounting and system status data are written into files which are periodically dumped on tape. There is no bias introduced in making these measurements, since the recording of these data is an integral part of the system operation. (Accounting tapes are used for billing and system status information is used to control certain dynamic scheduling parameters.)
5.
Configuration The basic system configuration during the periods in which measurements were taken is shown in Figure i. Hardware includes a duplex IBM 360/67 with two megabytes of core storage, two channel controllers, two paging drums, and numerous disks, tapes, and terminal controllers. The software controlling this equipment includes a time sharing supervisor known as UMMPS which implements a multi-tasking environment. The Michigan Terminal System (Mrs) runs as a task under UMMPS and copies of MrS provide service to both interactive and batch jobs. Spooling services for batch jobs
13
a r e p r o v i d e d by a v e r s i o n o f HASP m o d i f i e d t o r u n u n d e r UMMPS. The o t h e r t a s k o f m a j o r i m p o r t a n c e i n t h e s y s t e m i s t h e P a g i n g Drum P r o c e s s o r (PDP). which o v e r s e e s p a g i n g o p e r a t i o n s . .
Workload Similar to those of other universities, t h e Computing C e n t e r o f t h e U n i v e r s i t y o f M i c h i g a n p r o c e s s e s a l a r g e number o f r e l a t i v e l y s h o r t s t u d e n t j o b s and a l e s s e r number o f l a r g e r c o m p u t i n g r e q u e s t s , u s u a l l y g e n e r a t e d by t h e r e s e a r c h community. A j o b i n t h i s c o n t e x t means a s i n g l e s e s s i o n on a t e r m i n a l ( f r o m dial-up to disconnect) or a single batch submission (card deck). (Although it is possible to submit batch jobs to the system via disk files during a t e r m i n a l s e s s i o n , such j o b s a c c o u n t f o r l e s s t h a n 5% o f a l l b a t c h j o b s . ) T y p i c a l l y , 100,000+20,000_ j o b s a r e r u n p e r month, w i t h i n t e r a c t i v e j o b s c o m p r i s i n g 50% t o 60% o f t h e s e . The w o r k l o a d does v a r y c o n s i d e r a b l y w i t h t h e t i m e o f d a y , as shown i n F i g u r e 2. S i n c e t h i s g r a p h p r i m a r i l y r e f l e c t s work h a b i t s o f u s e r s , i t s s h a p e i s q u i t e s t a b l e from month t o month.
.
The Data F i g u r e s 3 and 4 summarize t h e mean CPU u t i l i z a t i o n and mean b a t c h queue l e n g t h by h o u r o f day d u r i n g t h e month o f O c t o b e r , 1973. (These t a b l e s a r e a c t u a l l y g e n e r a t e d f o r 2 4 - h o u r d a y s , b u t have b e e n t r u n c a t e d t o f i t s p a c e l i m i t a t i o n s . ) An e x a m i n a t i o n o f t h e s e summaries i n d i c a t e s t h a t t h e s y s t e m e x i s t e d i n t h r e e roughly definable operating regions: i.
Lighly loaded:
0% g CPU < 60%; Batch Queue = 0
2.
Moderately loaded:
3.
Heavily loaded:
60% v CPU < 90%; Batch Queue g 5
90% ~ CPU; Batch Queue > S
The u n d e r l i n e d p e r i o d s were c h o s e n f o r more d e t a i l e d s t u d y on t h e b a s i s o f these categories. To m i n i m i z e t r a n s i e n t e f f e c t s and s t a r t u p and shutdown p r o b l e m s , m e a s u r e m e n t p e r i o d s were r e q u i r e d t o be o f a t l e a s t two h o u r s d u r a t i o n , d u r i n g which t i m e a l l m a j o r h a r d w a r e (CPU's, s t o r a g e , and c h a n n e l c o n t r o l l e r s ) was i n o p e r a t i o n and t h e System was i n a s t a b l e s t a t e . D e s i r e d d a t a were e x t r a c t e d from t h e s y s t e m s t a t u s and a c c o u n t i n g t a p e s f o r t h e m e a s u r e m e n t p e r i o d s , and a r e p r e s e n t e d i n F i g u r e S. The p e r i o d s h a v e b e e n o r d e r e d by i n c r e a s i n g CPU u t i l i z a t i o n . Statistics presented include (1) t h e r a t i o o f b a t c h t o t e r m i n a l j o b s p r o c e s s e d , (2) a c t u a l CPU u t i l i z a t i o n , (5) t h e p e r c e n t a g e o f t h e l a t t e r n o t c h a r g e d t o u s e r s ( t h i s r e p r e s e n t s CPU t i m e consumed by " o v e r h e a d " t a s k s , p r i m a r i l y HASP and t h e PDP), (4) mean CPU c o n s u m p t i o n c h a r g e d t o b a t c h and i n t e r a c t i v e j o b s , and (S) mean and s t a n d a r d d e v i a t i o n o f t h e b a t c h queue l e n g t h , a f t e r a d j u s t m e n t s . Adjustments to the b a t c h queue l e n g t h m e a s u r e m e n t s were r e q u i r e d s i n c e t h e r e c o r d e d q u e u e l e n g t h s sometimes i n c l u d e d j o b s which were n o t a v a i l a b l e f o r e x e c u t i o n . A number creases, mean CPU CPU t i m e
o f t r e n d s may be o b s e r v e d i n t h e d a t a . As t h e CPU u t i l i z a t i o n int h e p r o p o r t i o n o f b a t c h j o b s i n t h e mix t e n d s t o d e c r e a s e , and t h e t i m e u s e d by a b a t c h job t e n d s t o i n c r e a s e . C o n v e r s e l y , t h e mean u s e d by i n t e r a c t i v e j o b s g e n e r a l l y d e c r e a s e s as s y s t e m l o a d i n c r e a s e s .
14
Also, proportionately more of the non-idle CPU time is spent performing overhead operations when the system is heavily loaded. Records of time charged to individual overhead tasks are not saved on tape, but several samples collected while the system was running indicated that typically 60% of the overhead time is consumed by the PDP and about 40% by HASP and a few miscellaneous tasks. There is, however, considerable variance in these percentages with changes in the system load. Figures 6 and 7 show histograms of CPU usage for batch and terminal jobs during two of the measurement periods. Although the means and variances of these distributions vary from period to period, the shapes of these histograms are typical. The batch queue length distribution showed more variation between periods. Figures 8-10 were obtained by summing the distributions for the light, moderate, and heavy periods respectively. For comparison, Figures 11-12 present data from October, 1974-one year after the original measurements. The average batch queue lengths indicate the increased load on the system. The CPU usage distributions for batch and terminal jobs summed over the entire month of October, 1974, are presented in Figures 13 and 14. Again, the shapes of these distributions are typical, although means vary from month to month. (During January-November 1974, mean CPU consumption by terminal jobs varied between 20 and 29 seconds. The corresponding figures for batch jobs varied between 7 and 25 seconds, with the higher figures generally occurring in the summer months when fewer short student jobs are run.) 6.
Postscript and Acknowledgments The University of Michigan Computing Center obtained a single processor IBM 370/168 CPU with two megabytes of storage in January, 1975; future statistics will reflect this increase in CPU power. Preliminary reports indicate that CPU utilization has dropped from near 100% to about 40% with the new CPU. The statistics collection facilities used in gathering the data were constructed by M. T. Alexander, D. W. Boettner, W. S. Gerstenberger and other members of the University of Michigan Computing Center staff, whose work is gratefully acknowledged.
Reference i.
Landwehr, C. E., Load Sharing in Computer Networks, A Queueing Model. MERIT Computer Network MCN-ii74-TR-18, Ann Arbor, Michigan, November, 1974.
15
-•-]i I
,i j,
,
,
[~ i
-
~
I
I
~ L L ~ I
¢Q ~ ¢ . . l l . l t . l l
[
I -
Fig. ]
S
o
U n i v e r s i t y of Z~.ich~_q;~n C o r ~ u t l n ~ ~iachinc Confiqu~,:~ti~n 16
Center
NItH,BI;R 0F ILII'('ii t,~ll TI I~H??41. U S I h S S!(;~II.D ON BY IS l l l h t l l l ; INII.I
